Skip to content

Research.

We publish measurements of how speech systems behave in hotel operations, with the method and the results, so that others can check them.

Questions: support@tryharlow.com

Featured
Research report · · Version 1.2

The Harlow Hospitality ASR Benchmark

Measuring speech-to-request reliability in hotel conditions

A benchmark for speech recognition in hotel operations. It measures how often a spoken staff request goes ahead with a wrong room, name, item, count or negation while nobody is told, and compares systems at matched repeat rates on synthetic hotel requests and on real accented speakers.

  • Speech recognition
  • Evaluation
Read the publication PDF, 27 pages, 1.3 MB
Silently wrong
Hotel requests · Synthetic speech, 2,920 clips
2.6%Harlow5.1%Best single recognizer
Real accented speakers · 77 speakers, 1,386 clips
3.4%Harlow4.5%Best single recognizer
Share of requests that go ahead with a wrong detail while nobody is told, at matched repeat rates. Key results of the report.

Publications

1 publication

Reports are released under CC BY 4.0 unless a section says otherwise. To ask about a report, its data or a replication, email support@tryharlow.com.